About Form Control 2

form-control-2 — VoiceForm Universal Voice-controlled universal form filling: **Core (FastAPI + LLM/NLU)** + **Browser Extension (MV3)** + **Android AccessibilityService**, with an optional desktop native daemon. 🗂 Backlog The full zero-to-MVP backlog lives in backlog/: **27 spec'd items** across 6 phases — each with file paths, APIs, interfaces, acceptance criteria, and explicit handoffs. **GitHub issues #1–#27** mirror the files 1:1, labeled by phase (phase-0…phase-5), priority (P0/P1/P2), agent track (agent-*), and critical-path. **Milestones:** Phase 0 — Infra · Phase 1 — Core NLU · Phase 2 — Browser · Phase 3 — Android · Phase 4 — Desktop · Phase 5 — MVP Gate. Critical path (do not split):** INFR-001 → CONTRACT-001 → CORE-001 → EXT-002 → EXT-003 → EXT-006 → INTE-001 → REL-001 MVP = gates G1 + G2 + G3 + G4** (voice input → parsed commands → browser form filled → Android form filled). G5 (desktop native) is stretch. Start here: backlog/README.md — see "Coding Agent Takeaway List" for your track. ✅ Implemented so far Phase 1 — Core NLU** (the brain) is in: transcript + form context → validated Command[], ambiguity handling, sessions and a bridge-agnostic normalizer. | Ticket | Delivered | Where | |---|---|---| | CORE-001 | POST /api/v1/parse — label-first resolution, clarification instead of guessing, no invented values | core/nlu/parser.py, core/prompts/ | | CORE-002 | POST…

Release archive